Anabelian Phenomena in Geometry and Arithmetic

Anabelian Phenomena in Geometry and Arithmetic Florian Pop, University of Pennsylvania PART I: Introduction and motivation The term anabelian was invented by Grothendieck, and a possible translation of it might be beyond Abelian. The corresponding mathematical notion of anabelian Geometry is vague as well, and roughly means that under certain anabelian hypotheses one has: Arithmetic and Geometry are encoded in Galois Theory It is our aim to try to explain the above assertion by presenting/explaining some results in this direction. For Grothendieck s writings concerning this the reader should have a look at [G1], [G2]. A) First examples: a) Absolute Galois group and real fields Let K be an arbitrary field, K a an algebraic extension, K s the separable extension of K inside K a, and finally G K = Aut(K s K) = Aut(K a K) the absolute Galois group of K. It is a celebrated well known Theorem by Artin Schreier from the 1920 s which asserts the following: If G K is a finite non-trivial group, then G K = GR and K is real closed. In particular, char(k) = 0, and K a = K[ 1]. Thus the non-triviality + finiteness of G K imposes very strong restrictions on K. Nevertheless, the kind of restrictions imposed on K are not on the isomorphism type of K as a field, as there is a big variety of isomorphy types of real closed fields (and their classification up to isomorphism seems to be out of reach). The kind of restriction imposed on K is rather one concerning the algebraic behavior of K, namely that the algebraic geometry over K looks like the one over R. b) Fundamental group and topology of complex curves Let X be a smooth complete curve over an algebraically closed field of characteristic zero. Then using basic results about the structure of algebraic fundamental groups, it follows that the geometric fundamental group π 1 (X) of X is isomorphic as a profinite group to the profinite completion Γ g of the fundamental group Γ g of the compact orientable topological surface of genus g. Hence π 1 (X) is the profinite group on 2g generators σ i, τ i (1 i g) subject to the unique relation i [σ i, τ i ] = 1. In particular, the genus g of the curve X is encoded in π 1 (X). But as above, the isomorphy type of the curve X, i.e., of the 1

object under discussion, is not seen by its geometric fundamental group π 1 (X) (which in some sense corresponds to the absolute Galois group of the field K). Precisely, the restriction imposed by π 1 (X) on X is of topological nature (one on the complex points X(C) of the curve). B) Galois characterization of global fields More than forty years after the result of Artin Schreier, it was Neukirch who realized (in the late 1960 s) that there must be a p-adic variant of the Artin Schreier Theorem; and that such a result would have highly interesting consequences for the arithmetic of number fields (and more general, global fields). The situation is as follows: In the notations from a) above, suppose that K is a field of algebraic numbers, i.e., K Q a. Then the Artin Schreier Theorem asserts that if G K is finite and non-trivial, then K is isomorphic to the field of real algebraic numbers R abs = R Q a. This means that the only finite non-trivial subgroups of G Q are the ones generated by the G Q -conjugates of the complex conjugation; in particular, all such subgroups have order 2, and their fixed fields are the conjugates of the field of real algebraic numbers. Now the idea of Neukirch was to understand the fields of algebraic numbers K Q a having absolute Galois group G K isomorphic (as profinite group) to the absolute Galois group G Qp of the p-adics Q p. Note that G Qp is much more complicated than G R. It is nevertheless a topologically finitely generated field, and its structure is relatively known, by work of Jakovlev, Poitou, Jannsen Wingberg, etc, see e.g. [J W]. Finally, Neukirch proved the following surprising result, which in the case of subfields K Q is the perfect p-adic analog of the Theorem of Artin Schreier: Theorem (See e.g. Neukirch [N1]). For fields of algebraic numbers K, K Q a the following hold: (1) Suppose that G K = GQp. Then K is the decomposition field of some prolongation of p to Q a. Or equivalently, K is G Q -conjugated to the field of algebraic p-adic numbers Q abs p. (2) Suppose that G K is isomorphic to an open subgroup of G Qp. Then there exists a unique K Q a as at (1) above such that K is a finite extension of K. The Theorem above has the surprising consequence that an isomorphism of Galois groups of number fields gives rise functorially to an arithmetical equivalence of the number fields under discussion. The precise statement is as follows: For number fields K, let P(K) denote the set of their places. Let Φ : G K G L be an isomorphism of Galois groups of number fields. Then a consequence of the above Theorem reads: Φ maps the decomposition groups of the places of K isomorphically onto the decomposition places of L. This bijection respects the arithmetical invariants e(p p), f(p p) of the places p p, thus defines an arithmetical 2

equivalence: ϕ : P(K) P(L). Finally, applying basic facts concerning arithmetical equivalence of number fields, one gets: In the above context, suppose that K Q is a Galois extension. Then K = L as fields. Naturally, this isomorphism is a Q-isomorphism. Since K Q is a normal extension, it follows that K = L when viewed as sub-extensions of fixed algebraic closure Q a. In particular, G K = G L as subgroups of G Q. Thus the open normal subgroups of G Q are equivariant, i.e., they are invariant under automorphisms of G Q. This lead Neukirch to the following questions: 1) Does G Q have inner automorphisms only? 2) Is every isomorphism Φ : G K G L as above defined by the conjugation by some element inside G Q? Finally, the first peak in this development was reached at the beginning of the 1970 s, with a positive answer to Question 1) by Ikeda [Ik] (and partial results by Komatsu), and the break through by Uchida [U1], [U2], [U3] (and unpublished notes by Iwasawa) showing that the answer to Question 2) is positive. Even more, the following holds: Theorem. Let K and L be global fields. Then the following hold: (1) If G K = GL as profinite groups, L = K as fields. (2) More precisely, for every profinite group isomorphism Φ : G K G L, there exists a unique field isomorphism φ : L s K s defining Φ, i.e., such that Φ(g) = φ 1 g φ for all g G K. In particular, φ(l) = K. And therefore we have a bijection: Isom fields (L, K) = Out prof.gr. (G K, G L ) This is indeed a very remarkable fact: The Galois theory of the global fields encodes the isomorphism type of such fields in a functorial way! Often this result is called the Galois characterization of global fields. We recall briefly the idea of the proof, as it is very instructive for the future developments. First, recall that by results of Tate and Shafarevich, we know that the virtual l-cohomological dimension vcd l (K) := vcd(g K ) of a global field K is as follows, see e.g. Serre [S1], Ch.II: i) If K is a number field, then vcd l (K) = 2 for all l. ii) If char(k) = p > 0, then vcd p (K) = 1, and vcd l (K) = 2 for l p. In particular, if G K = GL then K and L have the same characteristic. Case 1. K, L Q a are number fields. Then the isomorphism Φ : G K G L defines an arithmetical equivalence of K and L. Therefore, K and L have the 3

same normal hull M 0 over Q inside Q a ; and moreover, for every finite normal subextension M Q of Q a which contains K and L one has: Φ maps G M isomorphically onto itself, thus defines an isomorphism Φ M : Gal(M K) Gal(M L) In order to conclude, one shows for a properly chosen Abelian extension M 1 M, every isomorphism Φ M which can be extended to an isomorphism Φ M1, can also be extended to an automorphism of Gal(M Q). Finally, one deduces from this that Φ can be extended to an automorphism of G Q, etc. Note that the fact that the arithmetical equivalence of normal number fields implies their isomorphism relies on the Chebotarev Density Theorem, thus analytical methods. Until now we do not have a purely algebraic proof of that fact. Case 2. K, L are global fields over F p. First recall that the space of all the non-trivial places P(K) of K is in a canonical bijection with the closed points of the unique complete smooth model X F p of K. In particular, given an isomorphism Φ : G K G L, the arithmetical equivalence of K and L, is just a bijection X 0 Y 0 from the closed points of X to the closed points of the complete smooth model Y F p of L. And the problem is now to show that this abstract bijection comes from geometry. The way to do it is by using the class field theory of global function fields as follows: First, one recovers the Frobenius elements at each place p of K; and then the multiplicative group K by using Artin s reciprocity map; and finally the addition on K = K {0}. Since the recipe for recovering these objects is invariant under profinite group isomorphisms, it follows that Φ : G K G L defines a group isomorphism φ K : K L. Finally, one shows that φ K respects the addition, by reducing it to the case φ K (x + 1) = φ K (x) + 1. Moreover, by performing this construction for all finite sub-extensions K 1 K of K s K, and the corresponding finite sub-extensions L 1 L of L s L, and using the functoriality of the class field theory, one finally gets a field isomorphism φ : K s L s which defines Φ, i.e., Φ(g) = φ g φ 1 for all g G K. PART II: Grothendieck s Anabelian Geometry The natural context in which the above result appears as a first prominent example is Grothendieck s anabelian geometry, see [G1], [G2]. We will formulate Grothendieck s anabelian conjectures in a more general context later, after having presented the basic facts about étale fundamental groups. But it is easy and appropriate to formulate here the so called birational anabelian Conjectures, which involve only the usual absolute Galois group. A) Warm-up: Birational anabelian Conjectures The so called birational anabelian Conjectures place the Results by Neukirch, Ikeda, Iwasawa, Uchida et al at least conjecturally into a bigger 4

picture. And in their most naive form, these conjectures assert that the isomorphy type of the absolute Galois group encodes the isomorphy type of a finitely generated infinite field up to a finite purely inseparable extension. Recall that for an arbitrary field K we denote by K i its maximal purely inseparable extension. Thus if char(k) = 0, then K i = K. Further, we say that two field homomorphisms φ, ψ : L K differ by an absolute Frobenius twist, if ψ = φ Frob n on L i for some power Frob n of the absolute Frobenius Frob. Birational anabelian Conjectures. (1) There exists a group theoretic recipe in order to recover finitely generated fields K from their absolute Galois groups G K. In particular, if for such fields K and L one has G K = GL, then K i = L i. (2) Moreover, given such fields K and L, one has the following: Isom-form: Every isomorphism Φ : G K G L is defined by a field isomorphism φ : L a K a, and φ is unique up to Frobenius twists. In particular, one has φ(l i ) = K i. Hom-form: Every open homomorphism Φ : G K G L is defined by a field embedding φ : L a K a, and φ is unique up to Frobenius twists. In particular, one has φ(l i ) K i. As in the case of global fields, the Isom-form of the Birational anabelian Conjecture is also called the Galois characterization of the finitely generated infinite fields. The main known facts are summarized below: Theorem. (1) (See Pop [P2], [P3]) There is a group theoretical recipe by which one can recover in a functorial way finitely generated infinite fields K from their absolute Galois groups G K. Moreover, this recipe works in such a way that it implies the Isom-form of the birational anabelian Conjecture, i.e., every isomorphism Φ : G K G L is defined by an isomorphism φ : L a K a, and φ is unique up to Frobenius twists. (2) (See Mochizuki [Mzk3], Theorem B) The relative Hom-form of the birational anabelian Conjecture is true in characteristic zero, which means the following: Given function fields K and L over Q, every open G Q -homomorphism Φ : G K G L is defined by a unique field embedding φ : L a K a, which in particular, maps L into K. We give here the sketch of the proof of the Isom-form. Mochizuki s Homform relies on his proof of the anabelian conjectures for curves over sub-p-adic fields, and we will say some words about that later on in the Lecture. The main steps of the proof are the following: 5

The first part of the proof consists in developing a higher dimensional Local Theory which, roughly speaking, is a direct generalization of Neukirch s result above concerning the description of the places of global fields. Nevertheless, there are some difficulties with this generalization, because in higher dimensions the finitely generated fields do not have unique normal (or smooth) complete models. Recall that a model X Z for such a field K is by definition a separated, integral scheme of finite type over Z whose function field is K. We will consider only quasi-projective normal models, maybe satisfying some extra conditions, like regular, etc. In particular, if X is a model of K, then the Kronecker dimension dim(k) of K equals dim(x) as a scheme. One has: K is a global field if and only if every normal model X of K is an open of either X K := Spec O K if K is a number field, or of the unique complete smooth model X K F p of K, if K is a global function field with char(k) = p. Further, there exists a natural bijection between the prime Weil divisors of X K and the non-archimedean places of K. The basic result by Neukirch [N1] can be interpreted as follows: First let us say that a closed subgroup Z G K is a divisorial like subgroup, if it is isomorphic to a decomposition group Z q over some prime q of some global field L. Note that the structure of such groups as profinite groups is known, see e.g., Jannsen Wingberg [J W]. Then the decomposition groups over the places of K are the maximal divisorial like subgroups of G K. This gives then the group theoretic recipe for describing the prime Weil divisors of X K in a functorial way. In general, i.e., if K is not necessarily a global field, there is a huge variety of normal complete models X Z of K. In particular, we cannot hope to obtain much information about a single specific model X of K, as in general there is no privileged model for K as in the global field case. (Well, maybe with the exception of arithmetical surfaces, where one could choose the minimal model, but this doesn t help much...) A way to avoid this is to consider in a first approximation the space of Zariski prime divisors D K of K. This is by definition, the set of all the discrete valuations v of K defined by the Weil prime divisors of all possible normal models X Z of K. A Zariski prime divisor v is called geometrical if char(k) = char(kv), or equivalently, if v is trivial on the prime field of K, and arithmetical otherwise. Clearly, arithmetical Zariski prime divisors exist only if char(k) = 0. If so, and if v is defined by a Weil prime divisor X 1 of some normal model X Z, then v is geometric if and only if v is a horizontal divisor of X Z. We denote by DK 1 the space of all geometrical Zariski prime divisors of K. For every Zariski prime divisor v D K of K, let Z v be the decomposition group of some prolongation v s of v to K s. We will call the totality of all the closed subgroups of the form Z v the divisorial subgroups of G K or of K. Finally, as above, a closed subgroup Z G K is called divisorial like subgroup, if it is isomorphic to 6

a divisorial subgroup of a finitely generated field L with dim(l) = dim(k). The main results of the Local Theory are as follows, see [P1]: a) For a Zariski prime divisor v, the numerical data char(k), char(kv), and dim(k) are group theoretically encoded in Z v ; in particular, whether v is geometric or not. Further, the inertia group T v Z v of v s v, and the canonical projection π v : Z v G Kv are also encoded group theoretically in Z v. In particular, the residual absolute Galois group G Kv at all the Zariski prime divisors v of K is group theoretically encoded in G K. b) Every divisorial like subgroup Z G K is contained in a unique divisorial subgroup Z v of G K. Thus the divisorial like subgroups of G K are exactly the maximal divisorial like subgroups of G K. And the space D K is in bijection with the conjugacy classes of divisorial subgroups of G K. The results from the local theory above suggest that one should try to prove the birational anabelian Conjecture by induction on dim(k). This is the idea for developing a Global Theory along the following lines: First, the Isom-form of the birational anabelian Conjecture for global fields, i.e., dim(k) = 1, is known; and we think of it as the first induction step. Now suppose that dim(k) = d > 1. By the induction hypothesis, suppose that the Isom-form of the birational anabelian Conjecture is true in dimension < d. Then one recovers the field K i up to Frobenius twists from G K along the following steps (and from this recipe it will be clear, what we do mean by a group theoretic recipe ). Step 1) Recover the cyclotomic character χ K : G K Ẑ of G K. The recipe is as follows: Since dim(kv) = dim(k) 1 < d, the cyclotomic character χ Kv is known for each geometric Zariski prime divisor v D 1 K. Therefore, the cyclotomic character χ v : Z v π v χ Kv GKv Ẑ is known for all v DK 1. On the other hand, using the higher dimensional Chebotarev Density Theorem, see e.g., Serre [S3], it follows that ker(χ K ) is the closed subgroup of G K generated by all the ker(χ v ), v DK 1. Thus χ K is the unique character χ : G K Ẑ which coincides with χ v on each Z v. Next let T K = lim m µ m be the Tate module of K. Denote by Ẑ (1) the adic completion of Z with respect to all integers m relatively prime to char(k), and fix an identification ı of these two G K -modules. The Kummer Theory gives a functorial completion homomorphism as follows: K j K K ˆδ H 1 (K, Ẑ (1)), 7

where functorial means that performing this construction for finite extensions M K inside K a K, we get corresponding commutative inclusion-restriction diagrams (which we omit to write down here). An essential point to make here is that the completion morphism j K : K K is injective. This follows by induction on dim(k) by using the following fact: Let K k be the function field of a geometrically irreducible complete curve X k. Then K /k is the group of principal divisors of X, thus a free Abelian group. And for K a number field, one knows that K /µ K is a free Abelian group. Step 2) Recover the geometric small sets of Zariski prime divisors. We will say that a subset D DK 1 of Zariski prime divisors is geometric, if there exists a quasi-projective normal model X k of K such that D = D X is the set of Zariski prime divisors of K defined by the Weil prime divisors of X. Here, k is the field of constants of K. It is a quite technical point to show by induction on d = td(k), that the geometric sets of Zariski prime divisors can be recovered from G K, see Pop [P3]. Next let D = D X be a geometric set of Zariski prime divisors. One has a canonical exact sequence 1 U D K Div(X) Cl(X) 0, where U D are the units in the ring of global sections on X, and the other notations are standard. Since the base field k is either finite or a number field, the Weil divisor class group Cl(X) is finitely generated. Thus if X is sufficiently small, then Cl(X) = 0. A geometric set of Zariski prime divisors D = D X will be called a small geometric set of Zariski prime divisors, if the adic completion Ĉl(X) is trivial. One shows that the small geometric sets of Zariski prime divisors can be recovered form G K, see loc.cit. In this process, one shows that the adic completion of the above exact sequence can be recovered form G K too: 1 ÛD K Div(X) = v Ẑ Ĉl(X) 0, Step 3) Recover the multiplicative group K inside K. Let D = D X be a small geometric set of Zariski prime divisors of K. The resulting exact sequence defined above becomes 1 ÛD K v Ẑ 0, as Ĉl(X) = 0. Next let v D be arbitrary. Then the group of global units U D is contained in the group of v-units O v. Thus the (mod m v ) reduction homomorphism p v : O v Kv is defined on U D. Using some arguments involving Hilbertian fields, one shows that there exist many v D such that U D as well as ÛD are actually mapped isomorphically into Kv, respectively Kv; and moreover, that inside Kv one has ( ) p v (U D ) = ˆp v (ÛD) Kv. On the Galois theoretic side, the reduction map p v is defined by the restriction coming from the inclusion Z v G K. And moreover, since U D is contained in the 8

v-units, it follows that under the restriction map K H 1 (Z v, Ẑ (1)), the image of ÛD is contained in the image of the inflation map inf(π v ) : Kv H 1 (Z v, Ẑ (1)) defined by the canonical projection π v : Z v G Kv. Finally, the recipe to recover K inside K is as follows. First, for each small geometric set of Zariski prime divisors D and v as above, U D is exactly the preimage of ˆp v (ÛD) Kv, by assertion ( ) above. Since K = D U D, when D runs over smaller and smaller (small) geometric sets of Zariski prime divisors, we finally recover K inside K. Step 4) Define the addition in K = K {0}. This is easily done using the induction hypothesis: Let namely x, y K be given. Then x + y = 0 iff x/y = 1, and this fact is encoded in K. Now suppose that x + y 0. Then x + y = z in K iff for all v such that x, y, z are all v-units one has: p v (x) + p v (y) = p v (z). On the other hand, this last fact is encoded in the field structure of G Kv, which we already know. Finally, in order to conclude the proof of the Isom-form of the birational anabelian Conjecture, we proceed as follows: Let Φ : G K G L be an isomorphism of absolute Galois group Φ : G K G L. Then the recipe of recovering the fields K and L are identified via Φ, and shows that the p-divisible hulls of K and L inside K = L must be the same, where p = char(k). This finally leads to an isomorphism φ : L a K a which defines Φ. Its uniqueness up to Frobenius twists follows from the fact that given two such field isomorphisms φ, φ, then setting φ := φ 1 φ we obtain an automorphism of K a which commutes with G K. And one checks that any such automorphism is a Frobenius twist. B) Anabelian Conjectures for Curves a) Étale fundamental groups Let X be a connected scheme endowed with a geometric base point x. Recall that the étale fundamental group π 1 (X, x) of (X, x) is the automorphism group of the fiber functor on the category of all the étale connected covers of X. The étale fundamental group is functorial in the following sense: Let connected schemes with geometric base points (X, x) and (Y, y), and a morphism φ : X Y be given such that y = φ x. Then φ gives rise to a morphism between the fiber functors F x and F y, which induces a continuous morphism of profinite groups π 1 (φ) : π 1 (X, x) π 1 (Y, y) in the canonical way. In particular, setting Y = X and y some geometric point of X, a path from between x and y, gives rise to an inner automorphism of π 1 (X, x). In other words, π 1 (X, x) is determined by X up to inner automorphisms. (This means that the situation is completely parallel to the one in the case of the topological fundamental group.) It is one of the basic properties of the étale fundamental group that it is invariant under radicial morphisms, in particular under purely inseparable covers and Frobenius twists. 9

Next let G be the category of all profinite groups and outer continuous homomorphisms as morphisms. The objects of G are the profinite groups, and for given objects G and H, a G-morphism from G to H is a set of the form Inn(H) f, where f : G H is a morphism of profinite groups, and Inn(H) is the set of all the inner automorphisms of H. Clearly, if f, g : G H differ by an inner automorphism of G, then Inn(H) f = Inn(H) g, thus they define the same G-homomorphism from G to H. Further, Inn(H) f is a G-isomorphism if and only if f : G H is an isomorphism of profinite groups. Therefore, viewing the étale fundamental group π 1 as having values in G rather than in the category of profinite groups, the relevance of the geometric points x vanishes. Therefore, we will simply write π 1 (X) for the fundamental group of a connected scheme X. In the same way, if S is a connected base scheme, and X is a connected S- scheme, then the structure morphism ϕ X : X S gives rise to an augmentation morphism p X : π 1 (X) π 1 (S). Thus the category Sch S of all the S-schemes is mapped by π 1 into the category G S of all the π 1 (S)-groups, i.e. the profinite groups G with an augmentation morphism pr G : G π 1 (S). Now let us consider the more specific situation when the base scheme S is a field, and the k-schemes X are geometrically connected. Denote X = X k k s the base to the separable closure of k (in some fixed universal field ), and remark that by the facts above one has an exact sequence of profinite groups of the form 1 π 1 (X) π 1 (X) G k 1. In particular, we have a representation ρ X : G k Out(π 1 (X)) = Aut G (π 1 (X)) which encodes most of the information carried by the exact sequence above. The group π 1 (X) is called the algebraic (or geometric) fundamental group of X. In general, little is known about π 1 (X), and in particular, even less about π 1 (X). Nevertheless, if X is a k-variety, and k C, then the base change to C gives a realization of π 1 (X) as the profinite completion of the topological fundamental group of X an = X(C). In terms of function fields, if X k is geometrically integral, one has the following: Let k(x) k(x) be the function fields of X X. Then the algebraic fundamental group π 1 (X) is (canonically) isomorphic to the Galois group of a maximal unramified Galois field extension K X k(x). Finally, we recall that π 1 (X) is a birational invariant in the case X is complete and regular. In other words, if X and X are birationally equivalent complete regular k-varieties, then π 1 (X) = π 1 (X ) and π 1 (X) = π 1 (X ) canonically. b) Étale fundamental groups of curves Specializing even more, we turn our attention to curves, and give a short review of the basic known facts in this case. In this discussion we will suppose 10

that X is a smooth connected curve, having a smooth completion say X 0 over k. We denote S = X 0 \X, and S = X 0 \X. We will say that X is a (g, r) curve, if X 0 has (geometric) genus g, and S = r. We will say that X is a hyperbolic curve, if its Euler Poincaré characteristic 2 2g r is negative. And we will say that a curve X as above is virtually hyperbolic, if it has an étale connected cover X X such that X is a hyperbolic curve in the sense above. (Note that every étale cover f : X X as above is smooth and has a smooth completion which is a (g, r )-curve with g g and r r deg(f) over some finite k k.) In the above notations, let X k be a (g, r) curve. Then a short list of the known facts about the algebraic fundamental group π 1 (X) is as follows. First, let Γ g,r be the fundamental group of the orientable compact topological surface of genus g with r punctures. Thus Γ g,r = < a 1, b 1,..., a g, b g, c 1,..., c r i [a i, b i ] c j = 1 > is the discrete group on 2g + r generators a 1, b 1,..., a g, b g, c 1,..., c r with the given unique relation. (The generators a i, b i, c j have a precise interpretation as loops around the handles, respectively around the missing points.) In particular, if r > 0, then Γ g,r is the discrete free group on 2g + r 1 generators. It is well known that Γ g,r is residually finite, i.e., Γ g,r injects into its profinite completion: Γ g,r Γ g,r. Finally, given a fixed prime number p, respectively arbitrary prime numbers l, we will denote by Γ g,r Γ g,r the maximal prime-p quotient of Γ g,r, and by Γ g,r Γ l g,r the maximal pro-l quotient of Γ g,r. Case 1. char(k) = 0. Using the remark above, in the case k C, it follows that π 1 (X) = Γ g,r via the base change X k C X. If κ(x) is the function field of X, then π 1 (X) is the Galois group of a maximal Galois unramified field extension K k(x) k(x). Moreover, the loops c j Γ g,r around the missing points x i X 0 \X are canonical generators of inertia groups T xi over these points in π 1 (X). In particular we have: a) X is a complete curve of genus g if and only if π 1 (X) has 2g generators a i, b i with the singe relation i [a i, b i ] = 1, provided X is not A 1 k. b) X is of type (g, r) with r > 0 if and only if π 1 (X) is a profinite free group on 2g + r 1 generators, provided X is not k-isomorphic to A 1 k or P1 k. Clearly, the dichotomy between the above subcases a) and b) can be as well deduced from the pro-l maximal quotient π l 1 (X) of π 1(X), by simply replacing profinite by pro-l. Further, the following conditions on X are equivalent: (i) X is hyperbolic 11

(ii) X is virtually hyperbolic. (iii) π 1 (X) is non-abelian, or equivalently, (iii) l π1 l (X) is non-abelian. Case 2. char(k) > 0. First recall that the tame fundamental group π1 t (X) of X is the maximal quotient of π 1 (X) which classifies étale connected covers X X whose ramification above the missing points x i X 0 \X is tame. We will denote by π1 t (X) the tame quotient of π 1 (X), and call it the tame algebraic fundamental group of X. Now the main technical tools used in understanding π 1 (X) and its tame quotient π1 t(x) are the following two facts: Shafarevich s Theorem. In the context above, set char(k) = p > 0, and denote by π p 1 (X) the maximal pro-p quotient of π 1 (X). Further let r X0 = dim Fp Jac X0 [p] denote the Hasse Witt invariant of the complete curve X 0. Then one has: (1) If X = X 0, then π p 1 (X) is a pro-p free group on r X 0 g generators. (2) If X is affine, then π p 1 (X) is a pro-p free group on ka generators. Let k be an arbitrary base field, and v a complete discrete valuation with valuation ring R = R v of k and residue field kv = κ. Let X k be a smooth curve which has a smooth completion X 0 k. We will say that X k has good reduction at v, if the following hold: X 0 k has a smooth model X 0,R R over R, and there exists an étale divisor S R R of X 0,R such that the generic fiber of the complement X 0,R \S R =: X R R is X. Now let X k be a hyperbolic curve having good reduction at v. In the notations from above, let X s κ be the special fiber of X R R. Then the canonical diagram of schemes X X R X s k R κ gives rise to a diagram of fundamental groups as follows: π1 t(x) πt 1 (X R) π1 t(x s) G k G t k G κ where π t 1 (X R) is the tame fundamental group of X R, i.e., the maximal quotient of π 1 (X) classifying connected covers of X R which have ramification only along S R and the generic point of the special fiber, and this ramification is tame, and G t k is the Galois group of the maximal tamely ramified extension of k. The fundamental result concerning the fundamental groups in the diagram above is the following: 12

Grothendieck s Specialization Theorem. In the context above, let X be a smooth curve of type (g, r). Further let R t be the extension of R to k t, and X R := X R R R t. Then one has: (1) π t 1 (X) πt 1 (X R) is surjective, and π t 1 (X R) π t 1 (X s) is an isomorphism. The resulting surjective homomorphism sp v : π t 1(X) π t 1(X s ) is called the specialization homomorphism of tame fundamental groups at v. In particular, π t 1 (X) is a quotient of Γ g,r in such a way that the generators c j are mapped to inertia elements at the missing points x i X 0 \X. (2) Further, let char(k) = p, and denote by π 1 (X) the maximal prime to p quotient of π 1 (X) (which then equals the maximal prime to p quotient of π1 t(x) too). Then π 1 (X) = Γ g,r, and sp v maps π 1 (X) isomorphically onto π 1 (X s). In particular, π 1 (X) depends on (g, r) only. Combining Shafarevich s Theorem and Grothendieck s Specialization Theorem above, we immediately see that the following facts and invariants of X k are encoded in π 1 (X): a) If l p, then π l 1 (X) = Γ l g,r, and π p 1 (X) = Γ p g,r. Therefore, p = char(k) can be recovered from π 1 (X), provided X is not P 1 k. a) t The same is true correspondingly concerning the tame fundamental group π t 1 (X), provided X is not isomorphic to A1 k of P1 k. b) X k is complete if and only if π p 1 (X) is finitely generated. b) t Correspondingly, X k is complete if and only if π1 l (X) not pro-l-free, provided X is not isomorphic to A 1 k. c) In particular, if X is complete, then π1 l (X) has 2g generators, thus g can be recovered from π1 l(x). Finally, concerning the virtual hyperbolicity of X we have the following: d) Every affine curve X is virtually hyperbolic. Remark. Clearly, the applicability of Grothendieck s Specialization Theorem is limited by the fact that one would need a priori criterions for the good reduction of the given curve X at the (completions of k with respect to the) discrete valuations v of k. At least in the case of hyperbolic curves X k such criteria do exist. The setting is as follows: Let X k be a hyperbolic curve, and let v be a discrete valuation of k. Let T v Z v be the inertia, respectively the decomposition, groups of some prolongation of v to k s. Recall the canonical projections π 1 (X) G k and π1 t(x) G k and the resulting Galois representations ρ X : G k Out(π 1 (X)) and ρx t : G k Out(π1 t (X)). Then one can characterize the fact that X has (potentially) good reduction at v as follows, see Oda [O] in the case of complete 13

hyperbolic curves, and by Tamagawa [T1] in the case of arbitrary hyperbolic curves: In the above notations, X k has good reduction at v if and only if the representation ρ t X is trivial on T v. The concrete picture of how to apply the above remark in studying fundamental groups of hyperbolic curves X k over either finitely generated infinite base field k or finitely generated fields over some fixed base field k 0 is as follows: Let X k be a smooth curve of type (g, r). Further let S be a smooth model of k over Z, if k is a finitely generated field, respectively over the base k 0 otherwise. For every closed point s S, there exists a discrete valuation v s whose valuation ring R s dominates the local ring O S,s, and having residue field κ vs = κ(s). Let us choose such a valuation v s. Then in the previous notations, X has good reduction at s if and only if ρ t X is trivial on the inertia group T s over the point s. Note that by the uniqueness of the smooth model X Rs R s in the case it does exist, the existence of such a good reduction does not depend on the concrete valuation v s used. One should also remark here, that in the context above, X k has good reduction on a Zariski open subset of S. This follows e.g., from the Jacobian Criterion for smoothness. Finally, we now come to announcing Grothendieck s anabelian Conjectures for Curves and the Section Conjectures. Let P be a property defined for some category of schemes X. We will say that the property P is an anabelian property, if it is encoded in π 1 (X) in a group theoretical way, or in other words, if P can be recovered by a group theoretic recipe from π 1 (X). In particular, if X has the property P, and π 1 (X) = π 1 (Y ), then Y has the property P. Examples: a) In the category of all the fields K, the property K is real closed is anabelian. This is the Theorem of Artin Schreier from above. b) In the category of all the smooth k-curves X which are not isomorphic to A 1 k, the property: X is complete and has genus g is anabelian. This follows from the structure theorems for the fundamental group of complete curves as discussed above. We will say that a scheme X is anabelian if the isomorphy type of X up to some natural transformations, which are not encoded in Galois Theory, can be recovered group theoretically from π 1 (X) in a functorial way; or equivalently, if there exists a group theoretic recipe to recover the isomorphy type of X, up to the natural transformations in discussion, from π 1 (X). Typical examples of such natural transformations which are not seen by Galois Theory are the radicial 14

covers and the birational equivalence of complete regular schemes. Concretely, for k-varieties X k with char(k) = p > 0, there are two typical radicial covers: First the maximal purely inseparable cover X i k i. And second, the Frobenius twists X(n) k of X k and/or X i (n) k i of X i k i obtained by acting by Frob n on the coefficients : X(n) := X Frob n k k. In the same way, if X k and Y k are complete regular k varieties which are birationally equivalent, then π 1 (X) and π 1 (Y ) are canonically isomorphic, but X and Y might be very different. A good set of examples of anabelian schemes are the finitely generated infinite fields, as we have seen in the previous section. Given such a field K, one has π 1 (Spec K) = G K, and by the birational anabelian Conjectures we know that K can be recovered from π 1 (K) in a functorial way, up to pure inseparable extensions and Frobenius twists. Anabelian Conjecture for Curves (absolute form) (1) Let X k be a virtually hyperbolic curve over a finitely generated base field k. Then X is anabelian in the sense that the isomorphism type of X can be recovered from π 1 (X) up pure inseparable covers and Frobenius twists. (2) Moreover, given such curves X k and Y l, one has the following: Isom-form: Every isomorphism Φ : π 1 (X) π 1 (Y ) is defined by an isomorphism φ : X i Y i, and φ is unique up to Frobenius twists. Hom-form: Every open homomorphism Φ : π 1 (X) π 1 (Y ) is defined by a dominant morphism φ : X i Y i, and φ is unique up to Frobenius twists. One could as well consider a relative form of the above conjecture as follows: Anabelian Conjecture for Curves (relative form) (1) Let X k be a virtually hyperbolic curve over a finitely generated base field k. Then X k is anabelian in the sense that X k can be recovered from π 1 (X) G k up to pure inseparable covers and Frobenius twists. (2) Moreover, given such curves X k and Y k, one has the following: Isom-form: Every G k -isomorphism Φ : π 1 (X) π 1 (Y ) is defined by a unique k i -isomorphism φ : X i (n) Y i for some n-twist. Hom-form: Every open G k -homomorphism Φ : π 1 (X) π 1 (Y ) is defined by a unique dominant k i -morphism φ : X i (n) Y i of some n-twist. C) The Section Conjectures Let X 0 k be an arbitrary irreducible k-variety, and X X 0 an open k- subvariety. Let x X 0 be a regular k i -rational point of X 0. Then choosing a system of regular parameters (t 1,..., t d ) at x, we can construct by the standard procedure a valuation v x of the function field k(x) of X with value group 15

v x (K ) = Z d ordered lexicographically, and residue field k(x)v x = κ(x), thus a subfield of k i. Let v be a prolongation of v x to k(x) s, and let T v Z v be the inertia, respectively decomposition group of v v x in G K. By general valuation theory, see e.g., Kuhlmann Pank Roquette [K P R], one has: T v has complements G v in Z v. And clearly, since k(x)v x = κ(x) k i, under the canonical exact sequence 1 T v Z v G k(x)vx = G k 1, every complement G v is mapped isomorphically onto G k = G k i. Therefore, the canonical projection ( ) pr k(x) : G k(x) G k has sections s v : G k G v G k(x) constructed as shown above. Moreover, let us recall that under the canonical projection G k(x) π 1 (X), the decomposition group Z v is mapped onto the decomposition group Z x of v in π 1 (X), and T v is mapped onto the inertia group T x of v in π 1 (X). And finally, any complement G v of T v is mapped isomorphically onto a complement G x of T x in Z x. Clearly G x G k isomorphically, thus ( ) pr X : π 1 (X) G k has a section s x : G k G x π 1 (X) defined via the k i -rational point x X(k i ). Moreover, I think it s instructive to remark that one has to distinct cases: a) Suppose that x X. Then T x = {1}, as the étale covers of X are not ramified over x. Therefore, Z x = G x. And in this case the sections of pr X of the form above build a full conjugacy class of sections. b) Next let x (X 0 \X)(k i ). Then T x {1} and G x Z x in general. Therefore, for a given k i -rational point x, there might exist several complements G x of T x in Z x, thus sections of pr X : π 1 (X) G k, which are not conjugate inside π 1 (X). Such sections are called sections at infinity (for the variety X). In the case of an arbitrary variety X k it is not known how to classify the conjugacy classes of such sections. But if X k is a curve, and char(k) = 0, then these sections are classified by H 1 (G k, Ẑ(1)). Section Conjectures. Let X k be a hyperbolic curve over a finitely generated infinite field k, and in the case char(k) > 0, suppose that X has no finite covers which are defined over a finite field. In the notations from above, the following hold: (1) Birational form: The sections of pr k(x) arise from k i -rational points of X 0 as indicated above. (2) Curve form: The sections of pr X arise from k i -rational points of X 0 of X as indicated above. 16

Concerning Higher dimensional anabelian Conjectures, there are only vague ideas. There are some obvious necessary conditions which higher dimensional varieties X have to satisfy in order to be anabelian (like being of general type, being K(π 1 ), etc.). Also, easy counter-examples show that one cannot expect a naive Hom-form of the conjectures. See Grothendieck [G2], and Ihara Nakamura [I N], Mochizuki [Mzk3], [Mzk4] for more about this. Remark (Standard reduction technique). Before going into the details concerning the known facts about the anabelian Conjectures for curves, let us set the technical frame for a fact used several times below. Let X k be a smooth curve over the field k. Suppose that k is either a finitely generated infinite field, or a function field over some base field k 0. Let S Z, respectively S k 0 be a smooth model of k. Next let X k be a hyperbolic curve, say with smooth completion X 0. Let π 1 (X) G k, respectively π1 t(x) G k be the corresponding canonical projections. Then choosing for each closed point s S a discrete valuation v s which dominates the local ring of s, we have the Oda Tamagawa Criterion (mentioned above) for deciding whether X k has good reduction at s. We also know, that X k has good reduction on a Zariski open subset of S. In particular, the Oda Tamagawa Criterion is a group theoretic criterion for describing the Zariski open subset of S on which X k has good reduction. Moreover, if s is a point of good reduction of X k, then Grothendieck s Specialization Theorem for π1 t gives a commutative diagram of the following form: π1 t(x k v ) Z v sp s π t 1 (X s ) pr v Gκ(s) where k v is the fixed field of Z v inside k s, thus it is a decomposition field over v. Therefore, from the data π t 1 (X) G k, one recovers in a canonical way its special fibers π t 1 (X s) G κ(s) at all the points s where X has good reduction. I) Tamagawa s Results concerning affine hyperbolic curves In this subsection we will sketch a proof of the following result by Akio Tamagawa concerning affine hyperbolic curves. Theorem. (See Tamagawa [T1]) (1) There exists a group theoretic recipe by which one can recover an affine smooth connected curve X defined over a finite field from π 1 (X). Moreover, if X is hyperbolic, then this recipe recovers X from π t 1 (X). Further, the absolute and the relative Isom-form of the anabelian conjecture for Curves holds for affine curves over finite fields; and its tame form holds for affine hyperbolic curves over finite fields. 17

(2) There exists a group theoretic recipe by which one can recover affine hyperbolic curves X k defined over finitely generated fields k of characteristic zero from π 1 (X). Further, the absolute and the relative Isom-form of the anabelian conjecture for Curves holds for affine hyperbolic curves over finitely generated fields of characteristic zero. The strategy of the proof is as follows: First consider the case when the base field k is finitely generated and has char(k) = 0. We claim that the canonical exact sequence ( ) 1 π 1 (X) π 1 (X) G k 1, is encoded in π 1 (X). Indeed, recall that the algebraic fundamental group π 1 (X) is a finitely generated normal subgroup of π 1 (X). Therefore, since G k has no proper finitely generated normal subgroups, see e.g. [F J], Ch.16, Proposition 16.11.6, it follows that π 1 (X) is the unique maximal finitely generated normal subgroup of π 1 (X). Thus the exact sequence above can be recovered from π 1 (X). Further, by either using the characterization of the geometric inertia elements in π 1 (X) given by Nakamura [Na1], or by using specialization techniques, one finally recovers the projection π 1 (X) π 1 (X 0 ), where X 0 is the completion of X. After having recovered the exact sequence ( ) above, one reduces the case of hyperbolic affine curves over finitely generated fields of characteristic zero to the π1 t -case of affine hyperbolic curves over finite fields. This is done by using the standard reduction technique mentioned above. Tamagawa also shows that the absolute form of the Isom-conjecture and the relative one are roughly speaking equivalent (using the birational anabelian Conjecture described at the beginning of Part II). We now turn our attention to the case of affine curves over finite fields, respectively the π1 t -case of hyperbolic curves over finite fields. Tamagawa s approach is a tremendous refinement of Uchida s strategy to tackle the birational case, i.e., to prove the birational anabelian Conjecture for global function fields. (Naturally, since π 1 (X) seems to encode much less information than the absolute Galois group G k(x), the things might/should be much more intricate in the case of curves.) A rough approximation of Tamagawa s proof is as follows. Let X k be an affine smooth geometrically connected curve, where k is a finite field with char(k) = p. As usual let X 0 be the smooth completion of X. Thus we have surjective canonical projections π 1 (X) π 1 (X 0 ) G k 1 and correspondingly for the tame fundamental groups. The first part of the proof consists in developing a Local Theory, which as in the birational case, will give a description of the closed points x X 0 in terms of 18

the conjugacy classes of the decomposition groups Z x π 1 (X) above each closed point x X. The steps for doing this are as follows: Step 1) Recovering the several arithmetical invariants. Here Tamagawa shows that the canonical projections π 1 (X) G k, thus π 1 (X), as well as π 1 (X) π t 1 (X) are encoded in π 1(X). Further, by a combinatorial argument he recovers the Frobenius element ϕ k G k. In particular, one gets the cyclotomic character of π 1 (X), and so one knows the l-adic cohomology of π 1 (X), as well as the Galois action of G k on the l-adic Galois cohomology groups of π 1 (X). The next essential remark is that after replacing X by some sufficiently general finite étale cover Y X, the completion Y 0 k of Y is itself hyperbolic. In particular, the l-adic Galois cohomology groups H i (π 1 (Y 0 ), Z l (r)) of π 1 (Y 0 ) are the same as the l-adic étale cohomology H i (Y 0, Z l (r)) of Y 0. Thus by the remarks above, one can recover the l-adic cohomology of Y 0 for every étale cover Y X having a hyperbolic completion Y 0. This is a fundamental observation in Tamagawa s approach, as it can be used in order to tackle the following problem: Which sections of the canonical projection pr X : π 1 (X) G k are defined by points x X 0 (k) in the way as described in the Section Conjecture? We remark that since G k = Ẑ is profinite free on one generator, one cannot expect that all such sections are defined by points as asked by the Section Conjecture. (Indeed, there are uncountable many such conjugacy classes of sections, thus too many in oder to be defined by points, even if X 0 has no k-rational points.) Here is Tamagawa s answer: Let s : G k π 1 (X) be a given section. For every open neighborhood U π 1 (X) of s(g k ), we denote by X U X the finite étale cover of X classified by U. First, since U projects onto G k, the curve X U k is geometrically connected. Further, we have in tautological way: π 1 (X U ) = U, and U := U π 1 (X) = π 1 (X U ). Let X U,0 be the smooth completion of X U. Then by Step 1), the canonical projection π 1 (X U ) π 1 (X U,0 ) can be recovered from U = π 1 (X U ), thus from π 1 (X) endowed with the section s : G π 1 (X). Now we remark that for U sufficiently small, the complete curve X U,0 is hyperbolic, both in the case X is affine, or if X was hyperbolic and we were working π t 1 (X). We set U 0 := π 1 (X U,0 ) and view it as quotient of π 1 (X U ), and U 0 = π 1 (X U,0 ). Since X U,0 is complete and hyperbolic, the l-adic cohomology group H i et(x U0, Z l (1)) equals the Galois cohomology group H i (π 1 (X U0 ), Z l (1)), thus the cohomology group H i (U 0, Z l (1)) for all i. 19

Finally, since the Frobenius element ϕ k G k is known, by applying the Lefschetz Trace Formula, we can recover the number of k-rational points of X U0 : X U0 (k) = 2 i=0 ( 1)i Tr(ϕ k ) H i (U 0, Z l (1)) In this way we obtain the technical input for the following: Proposition. Let s : G k π 1 (X) be a section of pr X : π 1 (X) G k. Then s is defined by a point of X 0 (k) if and only if for every open sufficiently small neighborhood U of s(g k ) as above, one has X U0 (k) /O. Step 2) Recover the decomposition groups Z x over closed points. This is done using the Proposition above. Actually, using the Artin s Reciprocity law, one shows that in the case of a complete hyperbolic curve, like the X U0 above, the set X U0 (k) is in bijection with the conjugacy classes of sections s defining points. And this gives a recipe to recover the points X 0 (k) which come from points in X U0 (k) for some U as above; thus finally for recovering all the points in X 0 (k). By replacing k by finite extension l k, one recovers in a functorial way X(l) too, etc. Thus finally one recovers the closed points x of X 0 as being in bijection with the conjugacy classes of decomposition groups Z x π 1 (X). Correspondingly the same is done for π t 1 -case. The second part of the proof is to develop a Global Theory, as done by Uchida in the birational case. Naturally, π 1 (X) endowed with all the decomposition groups over the closed points of X 0 carries much less information than G k(x) endowed with all the decomposition groups Z v over the places v of k(x). Step 3) Recover the multiplicative group k(x) together with the valuations v x : k(x) Z. Since the Frobenius elements ϕ x Z x are known for closed points x X 0, by applying global class field theory as in the birational case, one gets the multiplicative group k(x) together with the valuations v x : k(x) Z. Step 4) Recovering the addition on k(x) {0}. This is much more difficult than that in the birational case. And here is were the hypothesis that X is affine is used. Namely, if x X 0 \X is any point at infinity, then from a decomposition group Z x over x, one finally can recover the evaluation map p x : k(x) k a. One proceeds by applying the following: Proposition. Let X 0 κ be a complete smooth curve over an algebraically closed field κ. Suppose that the multiplicative group k(x 0 ) together with the valuations v x : k(x 0 ) Z at closed points x X 0, and the evaluation of the functions at at least three k-points x 0, x 1, x of X are known. Then from these data the structure field of k(x 0 ) can be recovered. 20